根据《WIRED》报导与 Meta 内部安全通知，该公司旨在训练人工智慧模型的「模型能力倡议」（Model Capability Initiative）收集了美国员工笔记型电脑上的敏感数据，却因安全漏洞使内部所有人皆可存取。泄漏的数据包含 45,000 个 Hive 资料表，内容涵盖员工的敲击键盘记录、滑鼠点击、萤幕截图、私人对话以及个人与绩效资料。

此监控专案自四月份实施以来即引发员工强烈反弹，超过 1,600 名员工签署内部联署书抗议，指责此举侵犯隐私并带来安全与法规风险。尽管执行长祖克柏等高层辩护称，AI 模型需要透过观察优秀员工的操作来学习使用软体，但此次安全事件无疑证实了员工此前的担忧。

事件发生后，Meta 技术长 Andrew Bosworth 坦承存取控制列表（ACL）设定错误并承诺调查，而发言人也证实已无限期暂停该数据收集计划。在经历裁员与 AI 转型的动荡后，此次安全漏洞恐将使 Meta 内部本已低落的员工士气雪上加霜。

According to internal security notices and employee sources, Meta's "Model Capability Initiative"—a data collection program designed to train AI models—suffered a security breach that made sensitive data collected from US employees' laptops accessible to anyone inside the company. The exposed data spanning 45,000 Hive tables included keystrokes, mouse clicks, screen captures, private conversations, and performance metrics.

The surveillance program has faced intense employee backlash since its April launch, with more than 1,600 employees signing an internal petition warning of security risks and privacy violations. Although executives like CEO Mark Zuckerberg defended the project as necessary for AI models to learn from human behavior, this security failure validated the employees' pre-existing concerns.

Following the incident, Meta's Chief Technology Officer Andrew Bosworth admitted to misconfigured access control lists (ACLs) and promised a thorough investigation, while a spokesperson confirmed the company is pausing the data collection program indefinitely. This security breach is likely to worsen the ongoing morale crisis among employees at Meta.